DBRA-models

Model Checkpoints

This repository provides model checkpoints used in our personalized federated learning and defense experiments. The current release includes four groups of models.

Directory Structure

.

├── 0.75delta-Models/

├── client_increase-Models/

├── finetuned_models/

└── centralized_models/

Description

• 0.75delta-Models/

  Federated learning model checkpoints trained on non-IID datasets generated with a Dirichlet distribution using delta = 0.75.

• client_increase-Models/

  Federated learning model checkpoints trained on the controlled client-increase setting, where class 0 samples are manually assigned to clients with a fixed increasing pattern.

• finetuned_models/

  Pruned and fine-tuned model checkpoints corresponding to the defense method used in the paper.

• centralized_models/

  Centrally trained baseline models, including models such as CNN, VGG, and ResNet trained on CIFAR-10 or CIFAR-100.

File Format

The model checkpoints are stored as .pth files. Each checkpoint may contain the model parameters and related training information, depending on the training script used.

Notes

This repository is currently under organization. File names, model accuracy, and detailed loading instructions may be updated in later versions.